Claude Code Cloud Sessions Are Out of Preview, With a Credit Deadline Attached
Claude Code cloud sessions left research preview on September 23. Here's how claude --cloud actually works, and the October 7 deadline on Anthropic's trial credit.
Claude Code cloud sessions left research preview on September 23, moving from an experimental option to a supported way to run Claude Code that doesn't depend on your laptop staying open. Anthropic's developer account announced the change directly, and existing Pro and Max subscribers got a one-time trial credit to go with it: $100 on Pro, $250 on Max, claimable through October 7.
The credit is the part with a clock on it. The underlying feature is the part worth understanding if you run Claude Code as part of how you build things.
What a cloud session actually is
A cloud session is a Claude Code session that runs on infrastructure Anthropic manages instead of on your own machine, according to Anthropic's own documentation. The difference from a normal session isn't the model or the tools, it's where the process lives. Close your laptop mid-task in a local session and the work stops. Close it mid-task in a cloud session and it keeps running, because the session was never tied to your machine in the first place.
You can start one from four places: the browser at claude.ai/code, the Code tab in the Claude mobile app, the desktop app by selecting Cloud instead of Local when starting a session, or the terminal with claude --cloud. Scheduled and triggered routines also run as cloud sessions by default. Each surface reaches the same session, so a task you start from your phone on a commute shows up the same way if you open claude.ai/code from a laptop later that day.
Availability is Pro, Max, and Team plans, plus Enterprise users with premium seats or Chat + Claude Code seats, per Anthropic's docs. It requires GitHub access, either through the Claude GitHub App or by running /web-setup to send your local gh CLI token to your Claude account, since cloud sessions need somewhere to clone code from and push branches to.
Starting one from the terminal
The most direct way in for anyone already living in a terminal is the flag itself:
claude --cloud "Fix the authentication bug in src/auth/login.ts"
This clones your current repository's GitHub remote at your current branch, not your local checkout, so uncommitted local work needs pushing first. The task then runs in the cloud while you keep working locally, and you can fire off several of these back to back. Each --cloud call is its own independent session, which is the actual mechanism behind "run three things in parallel": three separate calls, three separate VMs, no shared state between them unless you build that yourself.
Getting a session back onto your machine works the other direction with --teleport, either picking from an interactive list or resuming a specific session ID. Teleport requires a clean git working directory, the right repository checkout, and the same claude.ai account that started the session. It's one-way: you can pull a cloud session down, but there's no equivalent command to push a running local session up. The Desktop app's Continue in menu covers that case instead.
The credit, and what it doesn't cover
The $100 Pro credit and $250 Max credit apply once per account, spent on cloud session usage before anything falls back to your normal plan limits, and multiple outlets covering the September 23 announcement (FrontierNews, AlphaSignal, dev.ua) converge on the same October 7 claim deadline and a November 4 expiration on unused balance. That timing detail didn't come from Anthropic's technical documentation, which doesn't mention promotional credits at all, so it's worth treating the deadline as reported rather than double-checking it against the docs page.
What the documentation is explicit about: there's no separate compute charge for a cloud VM. Cloud sessions share rate limits with every other kind of Claude and Claude Code usage on your account, and running several tasks in parallel draws down that shared limit proportionately faster, not from some separate cloud-specific pool. A free trial credit on top of that is a way to try the feature without touching your regular usage window, not a permanent second allowance.
How this is different from Claude Code Projects
Anthropic shipped a related feature, Claude Code Projects, on September 17, and it's easy to conflate the two since both revolve around sessions that survive a closed laptop. Projects is a coordination layer: one ongoing conversation that decides what becomes a new task and spins up a cloud session per thread, tracking all of them from one place. Cloud sessions are the substrate underneath that, the actual mechanism that lets a thread keep running unattended, and they exist independently of Projects. You can start a single claude --cloud task with no project involved at all, and plenty of real work looks exactly like that: one bounded fix, sent to the cloud, checked on from your phone an hour later.
What September 23 actually changed is that this stopped being a preview-gated experiment and became a standard way to run Claude Code, available to anyone on a Pro, Max, or Team plan rather than a limited testing group.
What it's actually good for
The clearest use case in Anthropic's own guidance is splitting planning from execution: run a local session in plan mode to work out the approach, commit the plan to the repo, then hand the actual work to a cloud session with claude --cloud "Execute the migration plan in docs/migration-plan.md". That keeps the judgment calls local and interactive while the mechanical part runs unattended.
The honest limitation is the same one that applies to any unattended agent: a cloud session waiting on your approval for an MCP tool call or a sign-in prompt counts as inactive, and inactive sessions eventually get their VM reclaimed, losing any background work like subagents or shell commands that was still running. For a bounded task with a clear finish line, that's a non-issue. For something open-ended that depends on you answering questions along the way, it's a reason to check in more than the "runs while you sleep" framing suggests.
For anyone running Claude Code as a real piece of their workflow rather than an occasional tool, the practical move is to use the credit window to actually test claude --cloud against a real task before October 7, not a toy one, since that's the version of the feature you'll be relying on either way once the trial balance runs out.
Sources: Claude Code cloud sessions documentation (Anthropic), ClaudeDevs announcement, credit and deadline details corroborated by FrontierNews and AlphaSignal.
Join the newsletter
AI workflows and systems, straight to your inbox.
No spam. Unsubscribe anytime.